Dstack

What is Dstack?

Dstack is an open-source orchestration engine that simplifies developing, training, and deploying AI models. It manages your clusters whether they run in the cloud or in your own data center, acting as a hub for AI engineers so you can handle development, training, and deployment without extra tools or waiting on an Ops team.

Compared to Kubernetes, Dstack is more lightweight and specialized for AI work. It works with multiple cloud providers, is easy to extend with support for new providers, and can manage on-prem servers as a group for running containers.

Who created Dstack?

Information about the creators of Dstack is not available in this source.

Key capabilities

• Multi-cloud and on-prem support: Connect Dstack to cloud accounts (AWS, Azure, GCP, and more) or your own data centers. Since it’s open-source and self-hosted, you can be up and running in minutes, using any hardware and open-source frameworks for training and deployment.
• Development environments: Create remote machines with your code and favorite IDE, letting you code interactively before deploying models or scheduling tasks.
• Tasks: Schedule batch jobs such as training, fine-tuning, or data processing on a single machine or across a cluster.
• Services: Deploy models as secure, public, and scalable endpoints.
• Fleets: Set up and manage clusters and instances in the cloud or on-premises, and reuse them across dev environments, tasks, and services.
• Dstack Sky: If you’d rather not host the Dstack server yourself, or want GPU access through a marketplace, Dstack Sky provides hosted servers and cloud GPUs at competitive prices.

With these capabilities, Dstack covers the full workflow of developing, training, and deploying AI models while managing clusters across cloud platforms and data centers.